3.5.2 PCIe AXI Slave IF Throughput
(Ask a Question)Throughput calculation is carried out at 250 MHz AXI CLK using fabric AXI DMA (64 KB of DMA size), maximum payload of 128 Bytes. The fabric AXI DMA 256-beat AXI burst length transactions on AXI slave IF. When using AXI slave IF, four PCIe outstanding transactions are supported.
| Link Width | Link Speed | PC to LSRAM (Memory Read from PC) | LSRAM to PC (Memory Write to PC) | Maximum Theoretical Throughput (MBps) | ||
|---|---|---|---|---|---|---|
| Maximum Throughput (MBps) | % of Theoretical Throughput | Throughput (MBps) | % of Theoretical Throughput | |||
| x1 | Gen1 | 204 | 81.6 | 226 | 90.4 | 250 |
| Gen2 | 411 | 82.2 | 453 | 90.6 | 500 | |
| x2 | Gen1 | 397 | 79.4 | 440 | 88 | 500 |
| Gen2 | 599 | 59.9 | 877 | 87.7 | 1000 | |
| x4 | Gen1 | 483 | 48.3 | 831 | 83.1 | 1000 |
| Gen2 | 665 | 33.25 | 1646 | 82.3 | 2000 | |
PCIe Throughput depends on the following factors:
- PCIe uses 8b10b encoding, which causes 20% reductions in throughput.
- The maximum PCIe read throughput also depends on the supported PCIe outstanding transactions and Round-trip time (RTT).
- Maximum effective bandwidth is the rate at which valuable data is transferred at a particular point. It does not include transaction overhead, such as headers, sequence numbers, CRCs, ECRCs, and other packets like DLLPs and SKIP advanced sets. Maximum Effective Bandwidth = data/(data + overhead)
The following table lists the relation between maximum transaction payload size and efficiency. Increasing the transaction payload size (increasing the burst length) also improves throughput.
| Maximum Transaction Payload Size (Byte) | Efficiency |
|---|---|
| 128 | 86% |
| 256 | 92% |
